Novelty Detection with MINAS

Clusters identified for different classes in stream.

This project is a Python implementation of the MINAS multi-class novelty detection algorithm for data streams. The implementation supports scikit-multiflow. The algorithm uses clustering techniques and has features for detecting novelties, concept extension, concept drift, and forgetting outdated stream states. In initial tests with the Kaggle credit card fraud detection data set, using prequential evaluation, the model achieved 82.7% recall for fraud examples.

Please check out more details and a quick-start guide on the project page.